We focus on developing robotic control and autonomous driving policy learning methods that could directly learn from real-world data, bypassing or alleviating sim-to-real gap, while achieving robust and generalizable performance.

Offline RL / IL / planning methods for autonomous driving and robotic control

Offline policy optimization for safety-critical scenarios

Foundation models for robotic control

Sim-to-real adaptation